Extinction-colonization dynamics structure genetic variation of spotted sunfish (Lepomis punctatus) in the Florida Everglades

Sample collection

In the Everglades ecosystem, spotted sunfish are relatively large (85 mm adult mean standard length [SL = length from the tip of the snout to the base of the caudal peduncle]), long-lived fish (sexual maturity at approximately 1 year); most Everglades fish are cyprinodontiformes with approximately annual life cycles (Loftus & Kushlan 1987; Trexler et al. 2001). We collected 278 spotted sunfish between 1996 and 1997 from eight canal and three marsh-pond sites distributed throughout Shark River Slough in Everglades National Park (ENP) and Water Conservation Areas 3A (WCA-3A) and 3B (WCA-3B) (Fig. 1; Table 1). Canal and levee constructions separate these three regions of the Everglades, with WCA-3B being completely isolated hydrologically from the rest of the ecosystem by a system of levees completed in 1963. Gates permit a regulated flow of water southward from WCA-3A to ENP. The sample size from each site ranged from 12 to 52, and the fish were obtained by angling and electrofishing. Upon collection, animals were transferred to the laboratory, standard length was measured, and they were stored at −80 °C prior to genetic analyses.

Sampling was conducted in the dry season of 1996 to obtain specimens from areas where they were concentrated because of declining water levels. However, an insufficient number of specimens was collected, so additional samples were obtained in the dry season of 1997 and pooled into a single sample with the fish from the previous year. We considered that pooling between the 2 years was appropriate because spotted sunfish live several years, permitting the adult fish from both years to be treated as members of the same cohort. Also, there was no evidence of significant interannual variation in allozyme or microsatellite allele frequencies in these two samples (Trexler et al. 2001).

We estimated the density of spotted sunfish from the marsh habitat adjacent to one of our marsh sampling sites in ENP (L35) from a long-term fish-sampling programme conducted by personnel of the park. These data were obtained by use of a 1-m² throw trap and approximately 105 samples were collected each year between 1978 and 1997. Fish were collected in February, April, July, October and December (for details of the sampling design see Loftus & Eklund 1994; Trexler et al. 2001). Juveniles of four species of sunfish found in the Everglades are not easily distinguishable (these are L. punctatus, L. marginatus, L. microlophus, and L. macrochirus), though only spotted sunfish and L. marginatus are abundant. We summed the counts of spotted sunfish and unidentified juvenile sunfish to yield a generous estimate of the historical density of spotted sunfish at the sampling site prior to our collection for genetic analysis.

Allozyme analysis

We used starch-gel electrophoresis to document patterns of allozymic variation in spotted sunfish. Whole-tissue extracts were prepared for electrophoresis by homogenization of tissues in approximately 500 µL of grinding buffer (0.025 m Tris pH 7.0, 0.025 m sucrose, 0.005 m mercaptoethanol). We pooled eye, liver and soma clips; preliminary analyses indicated no tissue-specific expression of the proteins, justifying this approach. We screened 32 protein-encoding loci with horizontal starch-gel electrophoresis before selecting 15 resolvable loci to score on all individuals. Of these, seven loci were polymorphic (Table 2). We followed standard techniques described in Selander et al. (1971) and Murphy et al. (1996), with 11% (w/v) starch gels. All allozymes were resolved using a Tris-citrate EDTA buffer at pH 8.0 (Ayala et al. 1972).

Microsatellite DNA analysis

We surveyed nine potential microsatellite-locus primer pairs in spotted sunfish that had been developed in closely related species, Lepomis auritus (DeWoody et al. 1998) and Lepomis macrochirus (Colbourne et al. 1996). From these nine loci, we identified five polymorphic loci that amplified reliably (Table 3). Whole genomic DNA was isolated from muscle tissue by standard phenol-chloroform DNA extraction methods (Hoelzel & Green 1992). The microsatellite loci were amplified using polymerase chain reaction (PCR) in 15 µL volumes, containing 1× buffer, 25 µm MgCl2, 250 µm dNTP, 5 U/µL Taq DNA polymerase (Promega), and a 5 µm primer set. One primer of each pair was end-labelled with a fluorescent dye (6-FAM, NED or HEX; Applied Biosystems). The thermal cycling parameters, modified from Colbourne et al. (1996), were as follows: an initial 1 minute denaturation at 94 °C, followed by 45 cycles of 30 s at 94 °C, 30 s at 55 °C, and 30 s at 72 °C, and a final 10 minute extension at 72 °C. We electrophoresed the amplified samples in 2.5% agarose gel to determine the presence or absence of a product. The amplified products were electrophoresed in 5% denaturing polyacrylamide gels on an ABI 377 automated DNA sequencer. The alleles were sized with respect to electrophoretic mobility compared to a ROX 350 standard and the genotypes were assigned using the genescan (ABI) and genotyper software packages.

Statistical methods

We examined patterns of allelic and genotypic diversity in both allozymes and microsatellites. We calculated goodness-of-fit to Hardy–Weinberg expectations at each allozyme and microsatellite locus within each site, and in global tests across loci and across sites, using genepop 3.2 (Raymond & Rousset 1995). When only two or three alleles were observed, we used the complete enumeration method of Louis & Dempster (1987), while we used the Markov chain method to calculate exact P-values for loci that exhibited more than three alleles (Guo & Thompson 1992). Linkage disequilibrium was tested for each pair of allozyme and microsatellite loci within each site using genepop 3.2 (Raymond & Rousset 1995). We used f-stat (Goudet 2001) to estimate gene diversity and allelic richness separately for allozymes and microsatellites. We also tested for differences in gene diversity and allelic richness between allozymes and microsatellites using a Mann–Whitney's U-test with loci as replicates.

We hypothesized that sunfish from marsh sites that experienced seasonal dry-down events that varied in severity might experience significant reductions in their effective population size. The most recent severe drought occurred in 1989–90, when all of ENP and WCA-3B, and most of WCA-3A were dry. Thus, we examined the possible impact of demographic changes from hydrologic fluctuations or habitat fragmentation on allozyme and microsatellite diversity. We estimated deviations from expected heterozygosity and tested for evidence of recent population bottlenecks with bottleneck (Cornuet & Luikart 1996; Piry et al. 1999). This program estimated heterozygosity excesses and deficiencies for each locus, and for each site, and tested significance using a one-tailed Wilcoxon test based on ranks. We used the infinite alleles model for analysis of the allozyme data, and both the stepwise and two-phase mutation models for analysis of microsatellites. In addition, an assignment test was performed according to the direct method (Rannala & Mountain 1997) with the leave-one-out option (geneclass software package; Cornuet et al. 1999). This permitted us to determine the likelihood that each individual's allozyme and microsatellite multilocus genotype belong to the site and/or region from which it was sampled (Paetkau et al. 1995). In order to increase the number of loci and thus the power of the population bottleneck and the assignment tests, allozyme and microsatellite data were combined, and the above analyses were repeated. If local or regional sites are stable and genetically distinct, then recent migrants to a region or a local site might be detected and the source of these migrants could be identified.

To test for isolation by distance and equilibrium of gene flow and drift, we calculated pairwise estimates of F_ST and R_ST for allozyme and microsatellite loci, respectively (Slatkin 1993; Hutchison & Templeton 1999). We tested the significance of pairwise correlations between these values and the linear geographical distance separating sites using the Mantel permutation procedure (Mantel 1967) with the Spearman's rank correlation coefficient as the test statistic (genepop 3.2; Raymond & Rousset 1995). Hydrographical distance between sites and linear geographical distance between sites were functionally equal because the Everglades is an expansive river-like wetland (i.e. with few exceptions, fish can move in all directions throughout much of the system). The patterns of the scatterplots were analysed according to Hutchison & Templeton (1999). Residual values from these analyses were classified as within or among regions, or within or among site types (canal or marsh). These values were analysed to test for regional division or site type patterns.

We used f-stat (Goudet 2001) to compare gene diversity and allelic richness separately for allozymes and microsatellites among regions and between marsh and canal sites. We used Weir & Cockerham's (1984) co-ancestry method to hierarchically partition the total observed genetic variation for the allozyme and microsatellite data. These partitions were attributable to variation among individuals within sites (individuals collected in areas < 1 km²) relative to total diversity within their site (θ_IS), among sites within water-management units relative to the total diversity in that unit (θ_SP), and among water-management units (regions) relative to the total genetic diversity (θ_ST). This provided estimates of Wright's hierarchical F-statistic of population subdivision based on a random-effects sampling model (Weir 1990). The regions correspond to an area enclosed by levees and canals: WCA-3A, WCA-3B, and ENP (Fig. 1). F-statistic estimates were also calculated separately among all canal sites and among all marsh sites to compare genetic structure estimates between deep-water refuge sites and sites that may be subject to frequent extinction and colonization. The Genetical Data Analysis program (gda 1.0; Lewis & Zaykin 1997) was used to estimate hierarchical F-statistics following Weir & Cockerham (1984), with bootstrapping across loci to estimate 95% confidence intervals.

To make statistical comparisons between the population-genetic patterns detected from allozymes and microsatellites, Weir & Cockerham's (1984) co-ancestry method was employed for both markers. Simulations performed by Ruzzante (1998) suggested that F_ST and R_ST perform similarly if mutation has not had enough time to affect allelic divergence of microsatellite loci. We tested for correlations of pairwise F_ST estimates between sites for allozymes and microsatellites with the Mantel permutation procedure performed as above.

Counts of sunfish collected in 1-m² samples deviated noticeably from a normal distribution, with many traps yielding zero specimens. Using SAS statistical software, we fit an overdispersed-Poisson linear model with a log linking function to provide estimates of annual density with asymmetrical confidence intervals lower bounded at zero (Littell et al. 1996).

Hardy–Weinberg equilibrium and genetic diversity

Genotypic frequencies of allozymes were more consistent with Hardy-Weinberg predictions than genotypic frequencies of microsatellites. The genotypic frequencies of the allozyme loci were generally consistent with expectations of Hardy-Weinberg equilibrium; testing across loci within each site, we found only one case that did not conform to Hardy-Weinberg equilibrium expectations (L1). Additionally, we found evidence of heterozygote deficiency (f < 0) at 3 of the 7 polymorphic loci examined for that site (Table 4). Several of the microsatellite loci did not conform to Hardy-Weinberg predictions. Testing within each site indicated that genotypic frequencies within three of the sites were inconsistent with Hardy-Weinberg predictions (L1, L11, and L32). Additionally, there was significant heterozygote deficiency detected for some loci within those sites, though different loci were deficient in heterozygotes among sites (Table 5). There was some evidence for linkage disequilibrium in allozymes (Gcdh & Pep) and in microsatellites (Lma29 & Lma87).

We observed moderate variation at the 7 polymorphic allozymes surveyed (7 of the 15 loci examined were polymorphic = 46.7%, average heterozygosity of loci that were polymorphic = 0.25; Table 2). The number of alleles per polymorphic locus across all sites ranged from 2 (Ldh, Idh-1) to 4 (Pgm-1 and Pep). We observed considerable variation in the five microsatellite loci studied (Table 5). The number of alleles per locus ranged from 14 (Lma21 and Lma87) to 17 (RB7 and Lma29; Table 5). The observed heterozygosity, averaged over all sites, ranged from 0.57 (Lma87) to 0.82 (RB7) (Table 3).

Compared to allozymes, microsatellites had a greater average gene diversity and allelic richness (Mann-Whitney's U = 75.00; P < 0.001). Matrices of allozyme and microsatellite pairwise F_ST estimates were significantly correlated (Mantel test, r = 0.571, P = 0.009), indicating concordance between the two types of molecular markers.

Recent bottleneck test

There were no indications that any of the studied sites experienced a recent, significant bottleneck. There was one site (L1) where the number of allozyme loci showing heterozygote excess was significantly greater than that expected from the allele diversity. Under the expectations of both models of mutation, there were no sites where the number of microsatellite loci showing heterozygote excesses was significantly greater than expected from the allele diversity. After applying a Bonferroni correction for multiple tests, none of these observations could be considered statistically significant (α = 0.05).

Assignment test

The assignment test performed with the allozyme data did not reliably assign individuals to their site or region of sampling origin. Only 57 of 231 (24.7%) were correctly assigned to the site from which they were sampled, and 80 of 231 (34.6%) correctly assigned to WCA-3A, WCA-3B or ENP. Following King et al. (2001), we found that the inclusion of all allozyme loci produced the greatest assignment accuracy. The assignment test performed with the microsatellite data did not reliably assign individuals to their site or region of sampling origin. Only 48 of 283 (17.0%) were correctly assigned to the site from which they were sampled, and 163 of 283 (57.6%) were correctly assigned to WCA-3A, WCA-3B or ENP. The inclusion and exclusion of microsatellite loci did not alter the assignment accuracy. Similarly, the assignment test performed with a pooled data set of allozymes and microsatellites did not reliably assign individuals to their site or region of sampling origin. Only 58 of 231 (25.1%) were correctly assigned to the site from which they were sampled, and 114 of 231 (49.4%) were correctly assigned to WCA-3A, WCA-3B, or ENP. This suggests that there is not a unique multilocus genotypic distribution for the allozyme or microsatellite loci that described a site or region. This could result from routine mixing within and/or between the regions. The assignment test assumes Hardy-Weinberg equilibrium, which was partially violated by the microsatellite data.

Isolation by distance

We found no correlations between pairwise estimates of geographical distance and F_ST from allozymes or F_ST and R_ST from microsatellite loci, further indicating a lack of spatially explicit genetic structure (Fig. 2). Furthermore, residual values calculated from these regressions did not reveal any significant patterns of deviation associated with region or site type. The overall pattern of the scatterplots was consistent between allozymes and microsatellites.

Hierarchical genetic diversity

Gene diversity and allelic richness were not significantly different between WCA-3A, WCA-3B and ENP (P > 0.05) for allozymes and microsatellites. They were not significantly different between marsh and canal sites for allozymes (P > 0.05). However, gene diversity and allelic richness were significantly different between marsh and canal sites for microsatellites (gene diversity; P = 0.022: allelic richness; P = 0.009). Marsh sites had, on average, less gene diversity and allelic richness than canal sites.

Allozymes and microsatellites revealed consistent and significant genetic structure among sites within regions (Table 6). When considered separately, sites within WCA-3A displayed significant genetic structure, but sites in WCA-3B and ENP did not (i.e. bootstrap confidence intervals cross zero). The F-statistic estimates were an order of magnitude larger for allozymes than for microsatellites. This is consistent with expectations of F-statistic estimates calculated from low to moderately variable loci compared to highly variable loci (reviewed in Hedrick 1999). Canal sites were not genetically differentiated, whereas marsh sites were significantly heterogeneous (confidence interval of θ_{PT marsh} did not cross zero). Based on a t-test re-sampling across loci with 1000 replicates, marsh sites were more heterogeneous than canal sites (H_O: θ_{PT canals} = θ_{PT marsh}; P = 0.115 for microsatelites and P = 0.109 for allozymes; Re-sampling test combining both data sets P = 0.08).

Spotted sunfish population dynamics

Spotted sunfish density in the marsh habitat near our ENP sampling sites fluctuated between approximately 0.5/m² to indistinguishable from zero over the 18 years prior to our study (Fig. 3a). No specimens of spotted sunfish (or of indistinguishable juveniles of the co-occurring congeners) were collected in 1988 or 1989, and very few specimens were collected in 1990. These latter low-density years overlapped with a relatively extreme 2-year drought event in 1989 and 1990. A caveat is required for the low estimate in 1988; that is the only year of the 2-year record where sampling events were missed but water levels were above the ground surface; no samples were collected in April or May of that year.

Hardy-Weinberg equilibrium

Some sites were deficient in the frequency of heterozygotes, especially for microsatellite loci. Heterozygote deficiencies can arise from inbreeding, underdominant selection, mixing of genetically differentiated populations at a site (spatial Wahlund effect), or mixing of cohorts (temporal Wahlund effect) (Hartl & Clark 1997). The concordance we observed between allozyme and microsatellite markers (pairwise F_ST Mantel test) suggests that gene flow and genetic drift were the major causes of any observed patterns of population differentiation (cf., Estoup et al. 1998; Ross et al. 1999). The absence of widespread patterns of heterozygote deficiency suggests that Wahlund effects are not principal factors in the genetic structuring of local populations of spotted sunfish.

The coefficient of variation (CV) of body size of individuals collected from site L1 was much greater than that of any other site sampled (CV_L1 = 8.91; CV_{OTHER SITES} = 2.76). The average size was not significantly different among sites with the exception of site L23 (Table 1). Although definitive discrimination between spatial and temporal Wahlund effects is beyond the power of our data, we pooled individuals into 1.0 cm size classes across all sites (6 size classes total, sample sizes within each class ranged from 14 to 39). We were unable to create size-based cohorts within sites because of low sample size. We re-tested for consistency with Hardy-Weinberg expectations within size classes for the microsatellites, and found only one locus (M87) that significantly deviated from expectations and only one of the cohorts was inconsistent with predicted frequencies (size class 6, 10.0 + cm, a mixture of larger/older classes). This suggests that temporal Wahlund effects should be considered in future studies and highlights the possibility that temporal variation and genetic differentiation between sites may be transient in time (Tessier & Bernatchez 1999). Genetic structure could be modified each generation by the addition of new cohorts and the mixing of old cohorts, depending on the connectivity of sites throughout the Everglades marsh. Thus, unique genetic signatures within sites may be masked by temporal variation (Whitlock & McCauley 1990; David et al. 1997); however, this does not preclude the possible effects of spatial mixing. It is interesting to note that pooling among sites within regions or among regions did not improve the fit of the data to Hardy-Weinberg expectations.

Patterns of genetic diversity

Allozyme and microsatellite loci differed in the amount of genetic diversity they showed, and in the patterns of gene diversity and allelic richness among sites. Highly variable microsatellite loci are likely more sensitive to these measures of genetic differentiation (Hedrick 1999). Both markers showed no evidence for recent population bottlenecks, and assignment tests did not reliably assign individuals to the site or region from which they were sampled. The assignment test failed to identify a unique genetic signal for regions and sites within regions. Thus, colonization of the marsh habitat after the dry season is not likely the result of a very small number of individuals that survive through the dry season in a deep-water refuge or colonize from canals to populate the site through in situ reproduction. Rather, the data indicate that a moderate to large immigrant population of spotted sunfish probably colonizes marsh sites. Also, a recent severe drought in 1989–90 did not cause a detectable bottleneck in those marsh sites, or subsequent colonization of the marsh sites from canal sites has masked the effects of the bottleneck.

We observed no evidence for isolation by distance in either allozymes or microsatellites, conforming to Hutchison & Templeton's (1999) case II. This pattern suggests that populations of spotted sunfish are not in equilibrium with respect to gene flow and genetic drift, possibly because of a recent colonization event from a relatively homogeneous source population. This conclusion is supported by the hierarchical population structure analyses, which did not reveal population structure for allozymes or microsatellites at the regional scale. Small but significant population structure and genetic differences were detected at the local level, which were likely the result of extinction and colonization effects at local spatial scales.

Local scale extinction-colonization dynamics of fish are tightly coupled with the natural hydrologic cycles of the Everglades (Trexler et al. 2001). We illustrated that the pattern with a 20-year record of spotted sunfish density in the marshes adjacent to one of our sites used for sampling genetic variation. Approximately 6 years prior to our study, the density of spotted sunfish dropped very low (indistinguishable from zero) for 1 to 2 years (1989 and 1990, we exclude the 1988 result because of low sample size). In 1990, alligator ponds in the vicinity of this marsh-sampling site dried, in addition to the marsh, because ground water depths dropped below 60 cm from the ground surface (Fig. 3b). We believe that this pattern of population dynamics is linked to our finding of modest population structure among marsh sites, but not among canal sites for spotted sunfish. This pattern is present even though the average distance between the marsh sites (16.8 km) was less than (P = 0.104) the average distance between canal sites (30.2 km). Such a pattern of population genetic structure is expected when a group of recently colonized populations is compared to populations that are not subject to extinction and colonization turnover (Whitlock & McCauley 1990). Additionally, the low F-statistics and small difference between canal and marsh F-statistics suggests that spotted sunfish metapopulations are intermediate along a gradient between propagule pool and migrant pool patterns (McCauley et al. 1995). One might also expect this result if ‘routine’ migration has diluted the effects of a population crash on patterns of genetic variation, possibly as in our study where the crash was noted in 1990, six or more years prior to our collections.

Our data suggest that canal sites yield a homogeneous pool of colonists for marsh sites that dry routinely. We hypothesize that marsh sites are colonized by a modest number of individuals from such sources representing a subset of the total genetic diversity present there, but not diminishing genetic variation enough to be detected statistically as a ‘bottleneck’. We propose that these colonization events create a modest genetic signature in the resulting marsh populations that may be short lived. Thus, a continent-island (canal-marsh) population structure best describes spotted sunfish genetics in the Everglades.

Genetic variation of Everglades spotted sunfish may not be at equilibrium at the regional level, in addition to the local dynamics already discussed. We do not know the historical population structure of sunfishes in the Everglades, but the regional habitat structure we sampled is a relatively recent addition to the Everglades ecosystem. The Tamiami Canal, separating WCA-3A and ENP, was originally constructed in 1926, and WCA-3A was enclosed by levees in 1963, thus the present regional habitat structure has been in place for 35–80 years. A minimum generation time for spotted sunfish is 2 years, so that 18–40 generations have passed since the regional structure was superimposed on the ecosystem. If the historical population was unstructured and the canals and levees form barriers to gene flow, the contemporary populations may not be at equilibrium with these changes. Genetic differentiation by drift, even with complete isolation (which is only possible for WCA-3B), is unlikely to have occurred in this short time without strong limits to dispersal and severe population bottlenecks (e.g. Arnaud et al. 2001).

The Everglades’ natural hydropattern has been disrupted by recent water management practices. Regions of the Everglades that were historically flooded only a fraction of the year, are now flooded for prolonged periods of time, while other areas are dry much longer each year than they were historically (Fennema et al. 1994). Also, historical patterns of sheet flow (flow of water across the wetland landscape driven by rainfall events) have largely been replaced with pulsed flow tied to management needs, resulting in lacustrine (southern WCA-3A) or stagnant conditions (WCA-3B) at some locations (Light & Dineen 1994; DeAngelis et al. 1997). Hydrologic patterns may affect the efficiency of colonization and exploitation of marsh habitats that become seasonally available to aquatic species in the Everglades. Our data suggest that patterns of periodic dry-down may create recurrent mixing events for spotted sunfish and, in this way, water management decisions may shape patterns of genetic diversity of this species in the contemporary Everglades ecosystem.